Modeling 3D Cardiac Contraction and Relaxation With Point Cloud Deformation Networks

Area of Science:

• Cardiovascular Imaging
• Medical Image Analysis
• Geometric Deep Learning

Background:

• Current cardiac function biomarkers like ejection fraction offer limited insight into the heart's 3D deformation.
• Understanding complex 3D cardiac mechanics is crucial for accurate diagnosis and improved patient outcomes.
• Existing methods struggle to capture the intricate details of biventricular anatomy deformation.

Purpose of the Study:

• To introduce the Point Cloud Deformation Network (PCD-Net), a novel geometric deep learning approach for direct 3D cardiac mechanics modeling.
• To analyze biventricular anatomy deformation throughout the cardiac cycle using point cloud representations.
• To evaluate PCD-Net's performance on a large dataset and its ability to detect myocardial infarction (MI).

Main Methods:

• Developed an encoder-decoder architecture, PCD-Net, utilizing point cloud deep learning for multi-scale feature extraction.
• Applied PCD-Net to a UK Biobank dataset of over 10,000 subjects for cardiac contraction and relaxation modeling.
• Utilized flexible and memory-efficient point cloud representations of cardiac anatomy.

Main Results:

• PCD-Net achieved predictive accuracy for cardiac deformation with Chamfer distances below image pixel resolution.
• The model successfully identified subpopulation-specific 3D cardiac mechanics differences between normal and myocardial infarction (MI) subjects.
• PCD-Net's 3D deformation encodings improved MI detection and prediction by 11% (AUC) and MI survival analysis by 7% (Harrell's C-index) compared to benchmarks.

Conclusions:

• PCD-Net offers a powerful new method for direct 3D cardiac mechanics modeling, surpassing traditional biomarkers.
• The network accurately captures complex cardiac deformations and identifies abnormal phenotypes associated with myocardial infarction.
• PCD-Net demonstrates significant potential for enhancing the diagnosis, prediction, and survival analysis of cardiovascular diseases like MI.